The local index formula in semifinite von Neumann algebras. II: the even case (Q2496243): Difference between revisions

For part I, see \textit{A.~L.\ Carey, J.~Phillips, A.~Rennie} and \textit{F.~A.\ Sukochev} [ibid.\ 202, 451--516 (2006; Zbl 1118.46060).] The main result obtained by Carey, Phillips, Rennie, and Sukochev (CPRS) in this excellent paper is the even local index theorem for semifinite even smooth spectral triples \((A, H, D)\) with spectral dimension \(q \geq 1\), where \(H\) is a Hilbert space, \(A\) is a \(\ast\)-subalgebra of a semifinite von Neumann algebra \(N\) acting on \(H\) and having a faithful normal semifinite trace \(\tau\) on \(N\), and \(D\) is a densely defined unbounded self-adjoint operator on \(H\) affiliated to \(N\) such that \([D, a]\) for \(a\in A\) is densely defined and extends to a bounded operator, \((\lambda - D)^{-1}\) for \(\lambda \not\in \mathbb R\) is compact with respect to \(\tau\) on \(N\), and, in addition, there exists a grading \(\gamma\in N\) such that \(\gamma^*=\gamma\), \(\gamma^2=1\), \(a \gamma -\gamma a =0\) for \(a \in A\) and \(D \gamma + \gamma D=0\), and the spectral dimension \(q\) of \((A, H, D)\) is defined to be the infimum of \(k\in \mathbb R\) such that \(\tau ((1 + D^2)^{-k/2} ) <\infty\). This is a sort of a generalization of the even local index theorem of \textit{A. Connes} and \textit{H. Moscovici} [Geom. Funct. Anal. 5, 174--243 (1995; Zbl 0960.46048)] but its starting point is the McKean-Singer formula for the index: for \(D\), \(N\), \(\gamma\) (unitary) as above and \(D^+ =(1-P)DP\), where \(P=(\gamma +1)/2\), the index of \(D^+\) with respect to \(\tau\) is equal to \(f(0)^{-1} \tau (\gamma f(D))\), where \(f\) is a continuous even function on \(\mathbb R\) with \(f(0)\not=0\) and \(f(D)\) is of trace class. The main result says that for \((A, H, D)\), \(D^+\) as above, the index of \(p D^+ p\) for \(p\) a projection of \(\mathfrak A\) is the residue of the even sum of the resolvent cocycles (that are double cocycles for \(A\) modulo holomorphic functions) of the Chern characters of \(p\), and the index is also computed by the residue of the even sum of sums of the zeta functions, and if \((A, H, D)\) has isolated spectral dimension, then the index is the even sum of the residue cocycles of the Chern characters of \(p\), where the residue cocycles (that are double cocycles for \(A\)) are defined by the residues of the resolvent cocycles, where by definition \((A, H, D)\) has spectral dimension \(q\) isolated if for an element \(b \in N\) of the form: \(a_0[D, a_1]^{(k_1)} \cdots [D, a_m]^{(k_m)} (1+D^2)^{-m/2- | k| }\) with \(T^{(n)} \equiv [D^2, [D^2, [\cdots, [D^2, T] \cdots ]]]\) (\(n\)-iterated) and \(| k| =k_1 + \cdots + k_m\) and \(a_j\in A\) (\(0 \leq j \leq m)\), the zeta functions \(\zeta_b(z-(1-q)/2) \equiv \tau (b (1+D^2)^{-z + (1-q)/2})\) have analytic continuation to a deleted neighbourhood of \(z=(1-q)/2\). Its corollary says that if \(1 \leq q <2\), then the assumption of isolated spectrum dimension is unnecessary. But it seems to be non-trivial to know what values those residues take. The whole of this paper is organized for proving the main result by Fredholm theory in semifinite von Neumann algebras: The local index theorem (involving Clifford-Bott periodicity, resolvent expansion, and pseudodifferential expansion), and the resolvent (and residue) cocycle(s), as sections.